Intra Switch Transport Protocol for Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current arbitration schemes in data switches require request-grant loops, leading to increased packet latency and congestion due to the need for acknowledgments before data transmission can begin, which is inefficient as port density increases and data flows compete for fabric resources.
Innovation Solution
Implementing a transport protocol that allows data packets to be sent from an ingress port to an egress port without initial acknowledgement, using a counter to manage transmission based on a default window size or bandwidth delay product, and adjusting acknowledgement rates to maintain output buffer occupancy, thereby reducing latency and congestion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If request-grant arbitration scheme is used, then data flow control is achieved, but packet latency increases and productivity decreases
Solution Approach 1:
The patent applies preliminary action by pre-configuring window sizes and arbitration parameters before data transmission begins. Each input port maintains a pre-established window size parameter that allows immediate packet transmission without waiting for dynamic arbitration decisions, thereby eliminating the request-grant loop latency while maintaining flow control through predetermined parameters.
Solution Approach 2:
The patent implements self-service by enabling each input port to autonomously manage its own transmission window based on pre-configured parameters. The input port independently tracks its transmission count and compares it against its allocated window size, eliminating the need for centralized arbitration and reducing latency while maintaining distributed flow control.
2Reliability
If acknowledgement is required before transmission, then congestion control is improved, but transmission time increases
Solution Approach 1:
The patent applies preliminary action by pre-establishing window sizes and transmission parameters before data flow begins. This allows packets to be transmitted immediately without waiting for acknowledgements, as the congestion control parameters are predetermined and automatically enforced through the window size mechanism, eliminating the time-consuming request-grant-acknowledge cycle.
Solution Approach 2:
The patent ensures continuity of useful action by allowing packet transmission to proceed continuously without interruption for acknowledgements. The window size mechanism maintains continuous congestion control enforcement while packets flow through the switch fabric without stopping for arbitration or acknowledgement cycles, thereby maximizing productivity.
3Productivity
If port density increases, then switch capacity improves, but congestion increases and reliability deteriorates
Solution Approach 1:
The patent applies local quality by assigning individual window sizes to each input port based on its specific traffic characteristics and fabric resource requirements. Each port receives a customized window size parameter that allows it to transmit at an appropriate rate without contributing to overall congestion, enabling high port density while maintaining congestion resistance through localized parameter optimization.
Solution Approach 2:
The patent utilizes parameter changes by dynamically adjusting window sizes assigned to different input ports based on fabric conditions and port-specific requirements. This parameter optimization allows the switch to accommodate increasing port density by allocating appropriate transmission parameters to each port, maintaining reliability and congestion control even as overall switch capacity increases.
Data Source
AI summary
In one implementation, a network device is configured to implement an intra switch transport protocol. The intra switch transport protocol is configured to determine a window size according to a predetermined bandwidth and a round trip time between a plurality of ingress ports and an egress port. The network device is configured to sending at least one data packet from an ingress port to an egress port without first receiving any acknowledgement from the egress port. The network device continues to send packets and increment a counter accordingly until the counter reaches the window size. The network device ceases transmission of subsequent data packets and/or queues subsequent data packets based on the counter exceeding the window size.